Introduction
DynamoAI offers 5 types of privacy penetration tests to evaluate model vulnerabilities related to the extractability of PII and other sensitive data from your model. This includes: PII Extraction, PII Reconstruction, PII Inference, Membership Inference and sequence extraction.
What is PII?
- When configuring a PII extraction, inference, or reconstruction attack, one of the most important hyperparameters to configure is the
attack.pii_classesparameter. This controls which types of PII the extraction attack will be run for. - The PII classes available to you will be dependent on the Named Entity Recognition (NER) model you choose to use, but if you go with the DynamoFL default NER model, the following classes are available to you:
| Class Name | Description | Example(s) |
|---|---|---|
| CREDIT_CARD | credit card number | "6504 8764 7593 8248" |
| CRYPTO_NUMBER | crypto wallet number (BTC) | "3ySdvCkTLVy7gKD4j6JfSaf5d" |
| DATE_TIME | date or time value | "September 1st", "Today", "12:00pm EST" |
| EMAIL_ADDRESS | email address | "[email protected]", "jane_doe [at] org [dot] com" |
| EVENT | named events (storms, battles) | "US Open", "Hurricane Katrina" |
| IBAN_CODE | International bank account number | "FR650154264610QJGP3UHAJDJ02" |
| IP_ADDRESS | IPv4/v6 address | "245.108.222.0", "269f:1230:73e3:318d:842b:daab:326d:897b" |
| LANGUAGE | language name | "Spanish" |
| LAW | law name | "GDPR" |
| LOCATION | address, geopolitical entity, facility -- country, street, buildings | "The United States", "Central Park", "123 Main St", "JFK" |
| MONEY | monetary values | "$10B contract" |
| NORP | affiliation-- ex. nationality, religion | "Democratic Party" |
| NUMBER | cardinal, numeric, ordinal value, quantity, percent | "10", "3rd", "10 kg", "20%” |
| ORGANIZATION | organization name -- ex. companies, agencies | "OpenAI", "OPEC", "SEC" |
| PASSPORT | passport number | "604876475", "Q24219489" |
| PERSON | person's name | "Eric", "Jane Doe", "Parker" |
| PHONE_NUMBER | phone number | "961-770-7727" |
| PRODUCT | objects, vehicles, foods, etc | "iPhone" |
| SSN_NUMBER | social security number | "865-50-6891" |
| WORK_OF_ART | name of work of art -- books, songs, etc | "The Great Gatsby" |